Local Principal Component Analysis Research Articles

Identification of polo-like kinase 1 (PLK1) in aggressive prostate cancer by paradigm analysis.

5006 Background: Integrative analysis that combines expression data, copy number variation, sequence, and epigenetic data with thousands of known biological interactions can help identify strong biological associations and novel “genotype to phenotype” associations. Methods: We performed multi-dataset differential expression analysis with Statistical Analysis of Microarrays (SAM) and gene set enrichment analysis (GSEA) across 10 publicly-available PCa microarray datasets to identify genes and pathways differentially expressed between local (n=471) and metastatic (n=220) PCa. We used PARADIGM, an integrated pathway analysis method, on PCa tumor samples with mRNA expression, copy number variation, and TMPRSS2:ERG fusion status data to infer differential fusion gene-related “activities” for pathway features (genes, protein complexes, etc) within a “Superimposed Pathway” representing a comprehensive collection of genetic interactions currently containing 20,314 known interactions among 16,362 concepts representing 6916 proteins, 7345 complexes, 1449 families, 55 RNAs, 15 miRNAs and 582 processes. Results: Out of the 7571 genes tested (those genes having data in two or more studies), the meta-analysis on gene expression revealed 210 (1.8% FDR) positively associated with PCa metastasis and 403 (0.94% FDR) negatively associated genes (threshold was p<0.001, 2-tailed Student’s t-test). GSEA highlighted cell proliferation, cell cycle control, and DNA damage repair pathways with metastatic tumors. The PARADIGM analysis identified a network containing 914 features connected by 1137 edges. PLK1 was identified as both highly expressed in metastatic PCa and as one of the fourteen hubs in the largest (596-feature) sub-network identified by PARADIGM. PLK1 was also associated with high Gleason Sum and recurrent disease in independent local PCa datasets. Conclusions: Using an approach pioneered by members of our SU2C/PCF supported PCa Dream Team, integrated analysis across multiple PCa datasets associates PLK1 activity with aggressive PCa and suggests it may provide a novel treatment target for at least a genetic sub-set of advanced PCa.

Generalized incremental manifold learning algorithm based on local smoothness

Most of the existing manifold learning algorithms are not capable of dealing with new arrival samples.Although some incremental algorithms are developed via extending a specified manifold learning algorithm,most of them have some disadvantages more or less.In this paper,a new and more Generalized Incremental Manifold Learning(GIML) algorithm based on local smoothness was proposed.GIML algorithm firstly extracted the local smooth structure of data set via local Principal Component Analysis(PCA).Then the optimal linear transformation,which transformed the local smooth structure of new arrival sample's neighborhood to its correspondent low-dimensional embedding coordinates,was computed.Finally the low-dimensional embedding coordinates of new arrival samples were obtained by the optimal transformation.Extensive and systematic experiments were conducted on both artificial and real image data sets.The experimental results demonstrate that the GIML algorithm is an effective incremental manifold learning algorithm and outperforms other existing algorithms.

Double monitoring of common and specific features for multimode process

ABSTRACTA novel monitoring framework for multimode processes is proposed in this article. Generally, although the different modes have different process characteristics, they still share some common features, and the specific information is the unique characteristic that is not shared by all modes. On the basis of principal component analysis (PCA), a new statistical method, termed as local principal component analysis (LPCA), is developed for modeling multimode process data. LPCA is a modified version of PCA with the idea of maximizing local data variance, which aims to discover some directions that data, from all modes, share comparatively similar variance structure. In this low‐dimensional representation, the common features are captured and the residual corresponding to each mode can be treated as specific features. With the obtained LPCA model, both the common and specific features are utilized for monitoring. Moreover, a Bayesian inference strategy is further adopted to derive global indices of specific features for fault detection. Its feasibility and validity of the proposed method are illustrated through a simple multivariate linear system and the Tennessee Eastman challenge problem. © 2013 Curtin University of Technology and John Wiley & Sons, Ltd.

Extracting respiratory signals from thoracic cone beam CT projections

The patient respiratory signal associated with the cone beam CT (CBCT) projections is important for lung cancer radiotherapy. In contrast to monitoring an external surrogate of respiration, such a signal can be extracted directly from the CBCT projections. In this paper, we propose a novel local principal component analysis (LPCA) method to extract the respiratory signal by distinguishing the respiration motion-induced content change from the gantry rotation-induced content change in the CBCT projections. The LPCA method is evaluated by comparing with three state-of-the-art projection-based methods, namely the Amsterdam Shroud method, the intensity analysis method and the Fourier-transform-based phase analysis method. The clinical CBCT projection data of eight patients, acquired under various clinical scenarios, were used to investigate the performance of each method. We found that the proposed LPCA method has demonstrated the best overall performance for cases tested and thus is a promising technique for extracting a respiratory signal. We also identified the applicability of each existing method.

New Intelligent Computer Intrusion Detection Method Using Hessian Local Linear Embedding and Multi-Kernel Support Vector Machine

Computer networks frequently collapse under the destructive intrusions. It is crucial to detection hidden intrusions to protect the computer networks. However, a computer intrusion often distributes high dimensional characteristic signals, which increases the difficulty of intrusion detection. Literature review indicates that limited work has been done to address the nonlinear dimension reduction problem in computer intrusion detection. Hence, this study has proposed a new intrusion detection method based on the Hessian Local Linear Embedding (HLLE) and multi-kernel Support Vector Machine (SVM). The HLLE was firstly used to reduce the dimension of the original intrusion date in a nonlinear manner. Then the SVM with multiply kernels was employed to detect the intrusions. A real computer network experimental system has been established to evaluate the proposed method. Four typical intrusions have been tested. The test results show high effectiveness of the new detection method. In addition, the new method has been compared with the single-kernel SVM with Local Linear Embedding (LLE) or Principal Component Analysis (PCA). The comparison results demonstrate that the proposed HLLE plus multi-kernel SVM can provide the best computer intrusion detection rate of 97.1%.

Solving Partial Differential Equations on Point Clouds

In this paper we present a general framework for solving partial differential equations on manifolds represented by meshless points, i.e., point clouds, without parameterization or connection information. Our method is based on a local approximation of the manifold as well as functions defined on the manifold, such as using least squares, simultaneously in a local intrinsic coordinate system constructed by local principal component analysis using $K$ nearest neighbors. Once the local reconstruction is available, differential operators on the manifold can be approximated discretely. The framework extends to manifolds of any dimension. The complexity of our method scales well with the total number of points and the true dimension of the manifold (not the embedded dimension). The numerical algorithms, error analysis, and test examples are presented.

A local mesh method for solving PDEs on point clouds

In this work, we introduce a numerical method to approximate differential operators and integrals on point clouds sampled from a two dimensional manifold embedded in $\mathbb{R}^n$. Global mesh structure is usually hard to construct in this case. While our method only relies on the local mesh structure at each data point, which is constructed through local triangulation in the tangent space obtained by local principal component analysis (PCA). Once the local mesh is available, we propose numerical schemes to approximate differential operators and define mass matrix and stiffness matrix on point clouds, which are utilized to solve partial differential equations (PDEs) and variational problems on point clouds. As numerical examples, we use the proposed local mesh method and variational formulation to solve the Laplace-Beltrami eigenproblem and solve the Eikonal equation for computing distance map and tracing geodesics on point clouds.

A novel matrix-based method for face recognition

Two-dimensional principal component analysis (2DPCA) is one of the representative techniques for image representation and recognition. However, it fails in detecting the local variation of images, which characterizes the most important modes of variability of face images. Motivated by the fact that the local spatial geometric structure of images is effectual in learning the representative image space, we assign different weight to each training image and then present a novel method, namely local two-dimensional principal component analysis (L2DPCA), which explicitly considers the variations among nearby data. Finally, we describe an effective algorithm L2DPCA+2DPCA to further reduce dimensionality reduction. Extensive experimental results on two-face databases (Yale and AR) show the efficiency of the proposed method.

Local and global principal component analysis for process monitoring

In this paper, a novel data projection method, local and global principal component analysis (LGPCA) is proposed for process monitoring. LGPCA is a linear dimensionality reduction technique through preserving both of local and global information in the observation data. Beside preservation of the global variance information of Euclidean space that principal component analysis (PCA) does, LGPCA is characterized by capturing a good linear embedding that preserves local structure to find meaningful low-dimensional information hidden in the high-dimensional process data. LGPCA-based T2 (D) and squared prediction error (Q) statistic control charts are developed for on-line process monitoring. The validity and effectiveness of LGPCA-based monitoring method are illustrated through simulation processes and Tennessee Eastman process (TEP). The experimental results demonstrate that the proposed method effectively captures meaningful information hidden in the observations and shows superior process monitoring performance compared to those regular monitoring methods.

MG-local-PCA method for reduced order combustion modeling

Chemistry tabulation techniques such as flamelet models are a popular way to account for detailed chemistry effects in numerical simulation. These techniques are based on the identification of a low dimensional manifold in chemical space that accurately represents chemical evolutions associated to a specific combustion regime. During the last years, several authors used the Principal Component Analysis (PCA) to identify low dimensional manifold for combustion problems. However, full coupling between this manifold and flow solver has not yet been performed to the authors knowledge. The present paper introduces a new approach called Manifold Generated by a Local PCA or MG-L-PCA, which fully couple the manifold identified by a PCA and a DNS flow solver. The first part of the paper presents the PCA approach. Then, the coupling between this manifold and a DNS solver is presented. The MG-L-PCA approach is finally validated against a DNS simulation of flame vortex interaction using both detailed mechanism and a FPI manifold. Unlike FPI, the MG-L-PCA reproduces the dispersion in the chemical space induced by the flame-vortex interaction both for the species and the source terms.

A facial sparse descriptor for single image based face recognition

Single image based face recognition under different variations such as occlusion, expression and pose has been recognized as an important task in many real-world applications. The popularly widely used holistic features are easily distorted due to occlusion and some other variations. In order to tackle this problem, the sparse local feature descriptor based recognition methods have become more and more important, and promising performance is obtained. The recently developed SIFT, which detects feature points sparsely and extracts feature locally for object matching between different views and scales, can also benefit single image based face recognition. However, we find in this paper that SIFT should not be directly used for face recognition, because face recognition differs from generic object matching. To this end, we develop a new framework for detecting feature keypoints sparsely, describing feature context and matching feature points between two face images. We call this new proposed framework as Facial Sparse Descriptor (FSD). Experiments are conducted to support our analysis of SIFT, and extensive experiments are also presented to validate the proposed FSD against SIFT and its two variants, two dense local feature descriptor (i.e., LBP and HoG), PCA and Gabor based methods on AR, CMU and FERET databases.

Some refinements of the standard autoassociative neural network

Improving the training algorithm, determining near-optimal number of nonlinear principal components (NLPCs), extracting meaningful NLPCs, and increasing the nonlinear, dynamic, and selective processing capability of the standard autoassociative neural network are the objectives of this article that are achieved independently by some new refinements of the network structure and the training algorithm. In addition, three different topologies of the network are presented, which make it possible to perform local nonlinear principal component analysis. Performances of all methods are evaluated by a stock price database that demonstrates their efficiency in different situations. Finally, as it will be illustrated in the last section, the proposed structures can be easily combined together, which introduce them as efficient tools in a wide range of signal processing applications.

Sharp feature extraction in point clouds

Sharp feature extraction has been playing an important role in point cloud processing. In this study, a novel method for extracting sharp features from point clouds is presented. It is proposed that in a given point cloud, the displacement between each of the points and the weighted average position in the given neighbourhood of that point is calculated, and the point is labelled as the candidate sharp feature point if the displacement is salient. The normal directions of the obtained candidate sharp feature points are estimated by means of local principal component analysis. Tensor voting is performed to refine the normal estimates. The displacement between a point and its locally weighted average position is projected along the estimated normal direction. The points with extreme projection values are defined as the final sharp feature points. The implementation of the proposed method on both synthesised and practical scanned point clouds show that the method is effective and robust for the purpose of sharp feature extraction.

Novel dynamic Bayesian networks for facial action element recognition and understanding

In daily life, language is an important tool of communication between people. Besides language, facial action can also provide a great amount of information. Therefore, facial action recognition has become a popular research topic in the field of human-computer interaction (HCI). However, facial action recognition is quite a challenging task due to its complexity. In a literal sense, there are thousands of facial muscular movements, many of which have very subtle differences. Moreover, muscular movements always occur simultaneously when the pose is changed. To address this problem, we first build a fully automatic facial points detection system based on a local Gabor filter bank and principal component analysis. Then, novel dynamic Bayesian networks are proposed to perform facial action recognition using the junction tree algorithm over a limited number of feature points. In order to evaluate the proposed method, we have used the Korean face database for model training. For testing, we used the CUbiC FacePix, facial expressions and emotion database, Japanese female facial expression database, and our own database. Our experimental results clearly demonstrate the feasibility of the proposed approach.

Computerized scar detection on renal cortical scintigraphy images

Renal cortical scintigraphy is a well-established functional imaging technique for visual analysis of radiopharmaceutical tracer distribution. However, the visual evaluation is subjective, causing interobserver variability, especially in a quantifiable number of scars. The purpose of this study was to develop new computerized methods in renal cortical scintigraphy image interpretation, particularly addressing activity distribution and cortex continuity (scars). The proposed methods involve preprocessing stages of model-based automatic kidney segmentation using active-shape model and image normalization (transforming each kidney image into a standardized image vector). For our previous computer-aided diagnosis scheme, two new image-based features [localized activity drop and principal component analysis (PCA)] were defined. Their performance was evaluated and compared with our previous scheme by using free-response receiver operating characteristic that is in terms of sensitivity (true-positive fraction) and the mean number of false positives (FPs) per image. Clinical tests were conducted in 297 patients (231 normal and 66 abnormal). The PCA-based image feature presented the best scar detection performance, followed by the localized activity drop feature. Both schemes were found to be superior to our previous computer-aided diagnosis scheme. In the PCA-based scheme, for sensitivity of 0.90 (76/84), the mean number of FPs was measured as 4.52 (1343/297). For another setting with reduced sensitivity of 0.79 (66/84), the mean number of FPs improved to 1.21 (360/297). Finally, a decision fusion scheme using 'majority voting' was also proposed, the sensitivity and mean number of FPs of which were measured as 0.83 (70/84) and 1.90 (563/297), respectively. The proposed methods have potential to provide effective second-reader information to nuclear medicine specialists in finding scar regions. Possible ways to improve the FP rate were also proposed.

Improving Performance and Accuracy of Local PCA

Local Principal Component Analysis (LPCA) is one of the popular techniques for dimensionality reduction and data compression of large data sets encountered in computer graphics. The LPCA algorithm is a variant of k-means clustering where the repetitive classification of high dimensional data points to their nearest cluster leads to long execution times. The focus of this paper is on improving the efficiency and accuracy of LPCA. We propose a novel SortCluster LPCA algorithm that significantly reduces the cost of the point-cluster classification stage, achieving a speed-up of up to 20. To improve the approximation accuracy, we investigate different initialization schemes for LPCA and find that the k-means++ algorithm [[AV07]] yields best results, however at a high computation cost. We show that similar ideas that lead to the efficiency of our SortCluster LPCA algorithm can be used to accelerate k-means++. The resulting initialization algorithm is faster than purely random seeding while producing substantially more accurate data approximation.

Visual Object Tracking Based on Combination of Local Description and Global Representation

This paper provides a novel method for visual object tracking based on the combination of local scale-invariant feature transform (SIFT) description and global incremental principal component analysis (PCA) representation in loosely constrained conditions. The state of object is defined by the position and shape of a parallelogram, which means that tracking results are given by locating the object in every frame using parallelograms. The whole method is constructed in the framework of particle filter which includes two models: the dynamic model and the observation model. In the dynamic model, particle states are predicted with the help of local SIFT descriptors. Local key point matching between successive frames based on SIFT descriptors provides us an important cue for the prediction of particle states; thus, we can efficiently spread particles in the neighborhood of the predicted position. In the observation model, every particle is evaluated by local key point-weighted incremental PCA representation, which can describe the object more accurately by giving large weights to the pixels in the influence area of key points. Moreover, by incorporating the dynamic forgetting factor, we can update the PCA eigenvectors online according to the object states, which makes our method more adaptable under different situations. Experimental results show that compared to other state-of-the-art methods, the proposed method is robust especially under some difficult conditions, such as strong motion of both object and background, large pose change, and illumination change.

Improving local PCA in pseudo phase space for fetal heart rate estimation from single lead abdominal ECG

This paper proposes an improved local principal component analysis (LPCA) in pseudo phase space for fetal heart rate estimation from a single lead abdominal ECG signal. The improved LPCA process can extract both the maternal ECG component and the fetal ECG component in an abdominal signal. The instantaneous fetal heart rate can then be estimated from the extracted fetal ECG waveform. Compared with the classical LPCA procedure and another single lead based fetal heart rate estimation method, our improved LPCA method has shown better robustness and efficiency in fetal heart estimation, testing with synthetic ECG signals and a real fetal ECG database from PhysioBank. For the real fetal ECG validating dataset of six long-duration recordings (obtained between the 22(nd) and 40(th) week of gestation), the average accuracy of the improved LPCA method is 84.1%.

Local matrix adaptation in topographic neural maps

The self-organizing map (SOM) and neural gas (NG) and generalizations thereof such as the generative topographic map constitute popular algorithms to represent data by means of prototypes arranged on a (hopefully) topology representing map. Most standard methods rely on the Euclidean metric, hence the resulting clusters tend to have isotropic form and they cannot account for local distortions or correlations of data. For this reason, several proposals exist in the literature which extend prototype-based clustering towards more general models which, for example, incorporate local principal directions into the winner computation. This allows to represent data faithfully using less prototypes. In this contribution, we establish a link of models which rely on local principal components (PCA), matrix learning, and a formal cost function of NG and SOM which allows to show convergence of the algorithm. For this purpose, we consider an extension of prototype-based clustering algorithms such as NG and SOM towards a more general metric which is given by a full adaptive matrix such that ellipsoidal clusters are accounted for. The approach is derived from a natural extension of the standard cost functions of NG and SOM (in the form of Heskes). We obtain batch optimization learning rules for prototype and matrix adaptation based on these generalized cost functions and we show convergence of the algorithm. The batch optimization schemes can be interpreted as local principal component analysis (PCA) and the local eigenvectors correspond to the main axes of the ellipsoidal clusters. Thus, this approach provides a cost function associated to proposals in the literature which combine SOM or NG with local PCA models. We demonstrate the behavior of matrix NG and SOM in several benchmark examples and in an application to image compression.

Investigation of the MILD combustion regime via Principal Component Analysis

Moderate or Intense Low oxygen Dilution (MILD) combustion is particularly appealing for coupling very high combustion efficiencies with very low pollutant emissions. Experimental and numerical studies have been devoted to MILD combustion. However, fundamental aspects of such combustion regime are still not completely understood and this generates uncertainties, especially from the modeling perspective. A novel methodology based on Principal Component Analysis (PCA) is proposed for the investigation of the main features characterizing the MILD combustion regime. High fidelity experimental datasets of CH 4/H 2 flames propagating into a hot and diluted co-flow are employed to this purpose. Results indicate that the proposed methodology can provide progress variables which coherently follow the modification of the flame structure experimentally observed at different levels of oxygen dilution in the co-flow. A local PCA approach is also proposed, based on the partition of the data sets into cluster, followed by the local application of PCA. Such approach ensures that the low-dimensional projection is characterized by the minimum parameterization error. Results indicate that local PCA leads to the identification of regions of the flow characterized by different physical processes, thus allowing the determination of optimal progress variables in each of them.